Nubank: Data-Driven Underwriting
Former Credit Director at Nubank
inpractise.com/articles/nubank-data-driven-underwriting
Former Credit Director at Nubank
Interview Transcript
This is a snippet of the transcript, sign up to read more.
Would you actually override the Serasa score or not?
It aligned with the product and growth strategy to incentivize and create virality within the product. In the same year, we also started looking at negative values of people with negative debt in the bureaus. Back then in Brazil, credit reports only had information about who consulted your information, like your CPF (Social Security number in Brazil), and whether you defaulted on a credit product. There wasn't a positive bureau, only a negative bureau.
This is a snippet of the transcript, sign up to read more.
Would you actually override the Serasa score or not?
We could collect data on whether you hadn't paid a telco bill, credit card, personal loan, mortgage, etc. We decided to apply the classic rule. If you have a negative mark, you have not paid anything or you're currently delinquent, we would block you from acceptance. That helped on reducing delinquency. Our policy for the first year was if you're doing 700, let's do 650, and see how it performs, and then we would wait a few months to see its performance.
This is a snippet of the transcript, sign up to read more.
Would you actually override the Serasa score or not?
The first model was launched in March 2016, primarily based on bureau information and KYC data shared by customers during the application process. This data was collected through the net process, combined with bureau information like credit reports and scores. Naturally, this model was much better than what we had before. However, since we applied features like high scores or the absence of negative marks, we had to maintain those filters. Machine learning models are excellent at interpolation but struggle with extrapolation. Removing those filters would mean the model hadn't encountered such customers before, making it difficult to accurately predict default rates for customers not part of the previously approved population. That was our situation in 2016.
This is a snippet of the transcript, sign up to read more.
Related Content
Copyright Notice
This document may not be reproduced, distributed, or transmitted in any form or by any means including resale of any part, unauthorised distribution to a third party or other electronic methods, without the prior written permission of IP 1 Ltd.
IP 1 Ltd, trading as In Practise (herein referred to as "IP") is a company registered in England and Wales and is not a registered investment advisor or broker-dealer, and is not licensed nor qualified to provide investment advice.
In Practise reserves all copyright, intellectual and other property rights in the Content. The information published in this transcript (“Content”) is for information purposes only and should not be used as the sole basis for making any investment decision. Information provided by IP is to be used as an educational tool and nothing in this Content shall be construed as an offer, recommendation or solicitation regarding any financial product, service or management of investments or securities. The views of the executive expressed in the Content are those of the expert and they are not endorsed by, nor do they represent the opinion of In Practise. In Practise makes no representations and accepts no liability for the Content or for any errors, omissions, or inaccuracies will in no way be held liable for any potential or actual violations of laws, including without limitation any securities laws, based on Information sent to you by In Practise.
© 2025 IP 1 Ltd. All rights reserved.